The upcoming Firefox 80 with VAAPI support; already available from ALA

25-ago-2020

As I run
sudo DOWNGRADE_FROM_ALA=1 downgrade firefox

I got the version which I currently have installed (the one marked with the + sign) and furthermore the important Firefox 80 which still I don’t have available (the one which I marked with the red highlighter)

Why this claiming?
Linux Hardware Reviews, Open-Source Benchmarks & Linux Performance - Phoronix Firefox 80 To Support VA-API Acceleration On X11 - Phoronix

Firefox 80 will include, finally, HW acceleration on X11!

Would be the case, so, to push Firefox 80 to all branches, also for the sake of bug haunting?

Since a long time, when I want to play HTML5 videos on Firefox on Linux, I rely on MPV [1] which gave, also on 4K videos, a really light usage of CPU;
(otherwise, when I play videos from Firefox itself - on Linux -, I got almost 80% - 100% CPU usage); furthermore I experienced, here and there, stuttering while scrolling (please notice that: I have “layers.omtp.enabled” set to true and also “gfx.webrender.all” set to true) [2]
in facts, when I run the bencharmk HTML5 Fish Bowl I always got 60 FPS. But without real HW acceleration: and Firefox eats a lot of CPU usage (100%)

[1] GitHub - MassimoGirondi/ff2mpv: A Firefox add-on for playing URLs in mpv.
[2] Firefox/Tweaks - ArchWiki

However: many thanks to Manjaro’s Team: I’m on the stable branch and for me is OK to wait :slight_smile:

2 Likes

Firefox is in Arch Stable, so it should jump into our unstable pretty soon. Also, I believe Firefox is fasttracked so it shouldn’t be long for it to get into our Stable repo.

2 Likes

When 80 will be stable?

When the Manjaro team determines it.

1 Like

Also there is firefox-developer-edition which is on version 80.0b8-1 at all branches. Not sure if that includes the vaapi support already though.

edit
regular FF is at 80 in unstable in the meanwhile

1 Like

In the meantime I installed firefox-developer-edition 80.0b8-1:
in about:config i set:

gfx.webrender.all > true
media.ffmpeg.dmabuf-textures.enabled > true
media.ffmpeg.vaapi-drm-display.enabled > true
media.ffmpeg.vaapi.enabled > true
media.ffvpx.enabled > false

And when playing streaming Full-HD videos, the CPU usage is dropped to 9-12% :+1:

EDIT:
I forget to mention the facts that I also had to export the following env variable:
MOZ_X11_EGL=1

2 Likes

This is probably preferable to grabbing from other repos.
(until the latest FF release is pushed to stable)

I can wait :slight_smile:
At least I reported that VA-API works as expected on my system (Ivy Bridge).

Heck I might argue that pacman -Uing from one of our Unstable mirrors would be preferable as well.
But still … we wait for the push … we usually have a good record on that expediency… lets see :blush:
(and who am I to judge … I left my ungoogled-chromium unbuilt for ~a week. :cry: … but its building now)

maybe go in testing or unstable for some users

Because FF 80.0 solves a security vulnerability marked as HIGH could it be fast-tracked?

https://www.mozilla.org/en-US/security/advisories/mfsa2020-36/

Just did that yesterday since I could not wait :wink:

sudo pacman -U https://manjaro.moson.eu/unstable/extra/x86_64/firefox-80.0-1-x86_64.pkg.tar.zst

I must say although it works fine with hw acceleration it still requires a lot more CPU power compared to mpv for example. I guess I still stick to mpv (triggered by a script) to watch stuff. Especially on BAT.

btw. v80 arrived at testing branch.

can confirm that VA-API is already supported in dev-edition and it works really good

1 Like

For me, the CPU usage on VA-API of Firefox is almost similar to mpv (mpv on 8%-10%), when playing videos; however I still get high CPU usage when scrolling pages: it go from 20% to 80%.
Can you also check this with htop?

With mpv I am at ~5% at minimal clock speeds. With FF it is around 10% and clockspeed is generally highter. I tested with a VP9 encoded video in 4k with 60 fps.

https://www.youtube.com/watch?v=6bRE0W5JHL8

And what about cpu usage when scrolling?
I also mean when scrolling fastly.

around 12 to 14 % (constantly scrolling up and down like crazy) :wink:
But ok, our CPU’s are totally different. 2c/4t vs 8c/16t

1 Like

Firefox is typically fast-tracked on Manjaro anyway. It barely got released yesterday so it should hit stable soon theoretically, if it follows suit to our other firefox fast-tracking.


Besides looking at CPU sensors, is there a way to verify that VA-API is working?

In terminal start FF: MOZ_LOG="PlatformDecoderModule:5" firefox
Then start a video and look at the output in the terminal. There should be some lines with VA-API…

You also need to set environment variable MOZ_X11_EGL=1 and set FF option:
Set media.ffvpx.enabled to false in about:config to disable the bundled FFmpeg.

Also your codec must be supported by the GPU for hw decoding. vainfo shows a list of supported codecs.

https://wiki.archlinux.org/index.php/Firefox#Hardware_video_acceleration

2 Likes

After installation, when I attempt to launch it, I got:

XPCOMGlueLoad error for file /usr/lib/firefox/libxul.so:
/usr/lib/libc.so.6: version `GLIBC_2.32' not found (required by /usr/lib/firefox/libxul.so)
Couldn't load XPCOM.

Which is the same as Anybody else having trouble launching new Firefox 80.0-1 on PBP? (but I don’t have a PineBook Pro; I am on X86-64 platform, stable branch).

However , as I’ve said, all is fine with firefox-developer-edition 80.0b8-1